Voice-Modul: Whisper STT + TTS Onyx fuer Hausmeister-Bot

homelab-ai-bot/telegram_bot.py

@@ -80,6 +80,7 @@ import requests as _req
 import llm
 import memory_client
 import monitor
+import voice
 from core import config
 
 logging.basicConfig(
@@ -383,6 +384,50 @@ async def handle_memory_callback(update: Update, ctx: ContextTypes.DEFAULT_TYPE)
             await query.edit_message_text("Fehler beim Loeschen.")
 
 
+async def handle_voice(update: Update, ctx: ContextTypes.DEFAULT_TYPE):
+    """Sprachnachricht: Whisper STT -> LLM -> TTS Antwort als Text + Sprache."""
+    if not _authorized(update):
+        return
+    voice_msg = update.message.voice
+    if not voice_msg:
+        return
+
+    await update.message.reply_text("🎙 Höre zu...")
+    try:
+        tg_file = await ctx.bot.get_file(voice_msg.file_id)
+        audio_data = await tg_file.download_as_bytearray()
+
+        text = voice.transcribe(bytes(audio_data))
+        if not text:
+            await update.message.reply_text("Konnte die Nachricht nicht verstehen.")
+            return
+
+        log.info("Voice transkribiert: %s", text[:100])
+        await update.message.reply_text(f"🗣 \"{text}\"\n\n🤔 Denke nach...")
+
+        channel_key = str(update.effective_chat.id)
+        session_id = memory_client.get_or_create_session(channel_key, source="telegram")
+
+        context.last_suggest_result = {"type": None, "candidate_id": None}
+        handlers = context.get_tool_handlers(session_id=session_id)
+        answer = llm.ask_with_tools(text, handlers, session_id=session_id)
+        if session_id:
+            memory_client.log_message(session_id, "user", text)
+            memory_client.log_message(session_id, "assistant", answer)
+
+        await update.message.reply_text(answer[:4000], reply_markup=KEYBOARD)
+
+        audio_out = voice.synthesize(answer[:4000])
+        if audio_out:
+            import io as _io
+            await update.message.reply_voice(voice=_io.BytesIO(audio_out))
+        else:
+            log.warning("TTS fehlgeschlagen — nur Text gesendet")
+    except Exception as e:
+        log.exception("Fehler bei Voice-Nachricht")
+        await update.message.reply_text(f"Fehler: {e}")
+
+
 async def handle_message(update: Update, ctx: ContextTypes.DEFAULT_TYPE):
     """Button-Presses und Freitext-Fragen verarbeiten."""
     if not _authorized(update):
@@ -517,6 +562,7 @@ def main():
     app.add_handler(CommandHandler("feeds", cmd_feeds))
     app.add_handler(CommandHandler("memory", cmd_memory))
     app.add_handler(CallbackQueryHandler(handle_memory_callback, pattern=r"^mem_"))
+    app.add_handler(MessageHandler(filters.VOICE, handle_voice))
     app.add_handler(MessageHandler(filters.TEXT & ~filters.COMMAND, handle_message))
 
     async def post_init(application):

homelab-ai-bot/voice.py

@@ -0,0 +1,76 @@
+"""Spracheingabe (Whisper) und Sprachausgabe (TTS) fuer den Hausmeister-Bot.
+
+Nutzt die OpenAI API direkt (nicht OpenRouter).
+"""
+
+import io
+import logging
+from typing import Optional
+
+import requests
+
+from core import config
+
+log = logging.getLogger("voice")
+
+_api_key: Optional[str] = None
+
+TTS_MODEL = "tts-1"
+TTS_VOICE = "onyx"
+WHISPER_MODEL = "whisper-1"
+
+
+def _get_key() -> str:
+    global _api_key
+    if not _api_key:
+        cfg = config.parse_config()
+        _api_key = cfg.raw.get("OPENAI_API_KEY", "")
+        if not _api_key:
+            log.error("OPENAI_API_KEY fehlt in homelab.conf")
+    return _api_key
+
+
+def transcribe(audio_bytes: bytes, filename: str = "voice.ogg") -> Optional[str]:
+    """Transkribiert Audio via Whisper API. Gibt Text zurueck oder None bei Fehler."""
+    key = _get_key()
+    if not key:
+        return None
+    try:
+        r = requests.post(
+            "https://api.openai.com/v1/audio/transcriptions",
+            headers={"Authorization": f"Bearer {key}"},
+            files={"file": (filename, io.BytesIO(audio_bytes), "audio/ogg")},
+            data={"model": WHISPER_MODEL, "language": "de"},
+            timeout=30,
+        )
+        if r.ok:
+            text = r.json().get("text", "").strip()
+            log.info("Whisper: %s", text[:80])
+            return text
+        log.warning("Whisper Fehler: %s %s", r.status_code, r.text[:200])
+    except Exception as e:
+        log.warning("Whisper Exception: %s", e)
+    return None
+
+
+def synthesize(text: str) -> Optional[bytes]:
+    """Erzeugt Sprache via TTS API. Gibt OGG-Bytes zurueck oder None bei Fehler."""
+    key = _get_key()
+    if not key:
+        return None
+    if len(text) > 4000:
+        text = text[:4000]
+    try:
+        r = requests.post(
+            "https://api.openai.com/v1/audio/speech",
+            headers={"Authorization": f"Bearer {key}", "Content-Type": "application/json"},
+            json={"model": TTS_MODEL, "input": text, "voice": TTS_VOICE, "response_format": "opus"},
+            timeout=30,
+        )
+        if r.ok:
+            log.info("TTS: %d bytes fuer %d Zeichen", len(r.content), len(text))
+            return r.content
+        log.warning("TTS Fehler: %s %s", r.status_code, r.text[:200])
+    except Exception as e:
+        log.warning("TTS Exception: %s", e)
+    return None

homelab.conf

@@ -176,6 +176,7 @@ FORGEJO_TOKEN="b874766bdf357bd4c32fa4369d0c588fc6193336"
 FORGEJO_SYNC_TOKEN="5402da0447b0eb6aede721a8748a08974ddc5c42"
 GITHUB_PAT="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"
 OPENROUTER_KEY="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"
+OPENAI_API_KEY="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"
 MEMORY_API_TOKEN="Ai8eeQibV6Z1RWc7oNPim4PXB4vILU1nRW2-XgRcX2M"
 MEMORY_API_URL="http://100.121.192.94:8400"